drivers/staging/lustre/lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/types.h>
  42 #include <linux/mm.h>
  43
  44 #include "../include/lustre/lustre_ioctl.h"
  45 #include "../include/lustre_lite.h"
  46 #include "../include/lustre_ha.h"
  47 #include "../include/lustre_dlm.h"
  48 #include "../include/lprocfs_status.h"
  49 #include "../include/lustre_disk.h"
  50 #include "../include/lustre_param.h"
  51 #include "../include/lustre_log.h"
  52 #include "../include/cl_object.h"
  53 #include "../include/obd_cksum.h"
  54 #include "llite_internal.h"
  55
  56 struct kmem_cache *ll_file_data_slab;
  57 struct dentry *llite_root;
  58 struct kset *llite_kset;
  59
  60 #ifndef log2
  61 #define log2(n) ffz(~(n))
  62 #endif
  63
  64 static struct ll_sb_info *ll_init_sbi(struct super_block *sb)
  65 {
  66         struct ll_sb_info *sbi = NULL;
  67         unsigned long pages;
  68         unsigned long lru_page_max;
  69         struct sysinfo si;
  70         class_uuid_t uuid;
  71         int i;
  72
  73         sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
  74         if (!sbi)
  75                 return NULL;
  76
  77         spin_lock_init(&sbi->ll_lock);
  78         mutex_init(&sbi->ll_lco.lco_lock);
  79         spin_lock_init(&sbi->ll_pp_extent_lock);
  80         spin_lock_init(&sbi->ll_process_lock);
  81         sbi->ll_rw_stats_on = 0;
  82
  83         si_meminfo(&si);
  84         pages = si.totalram - si.totalhigh;
  85         lru_page_max = pages / 2;
  86
  87         sbi->ll_cache = cl_cache_init(lru_page_max);
  88         if (!sbi->ll_cache) {
  89                 kfree(sbi);
  90                 return NULL;
  91         }
  92
  93         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
  94                                            SBI_DEFAULT_READAHEAD_MAX);
  95         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
  96         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  97                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  98
  99         ll_generate_random_uuid(uuid);
 100         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 101         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 102
 103         sbi->ll_flags |= LL_SBI_VERBOSE;
 104         sbi->ll_flags |= LL_SBI_CHECKSUM;
 105
 106         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 107
 108         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 109                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 110                                pp_r_hist.oh_lock);
 111                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 112                                pp_w_hist.oh_lock);
 113         }
 114
 115         /* metadata statahead is enabled by default */
 116         sbi->ll_sa_max = LL_SA_RPC_DEF;
 117         atomic_set(&sbi->ll_sa_total, 0);
 118         atomic_set(&sbi->ll_sa_wrong, 0);
 119         atomic_set(&sbi->ll_agl_total, 0);
 120         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 121
 122         /* root squash */
 123         sbi->ll_squash.rsi_uid = 0;
 124         sbi->ll_squash.rsi_gid = 0;
 125         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 126         init_rwsem(&sbi->ll_squash.rsi_sem);
 127
 128         sbi->ll_sb = sb;
 129
 130         return sbi;
 131 }
 132
 133 static void ll_free_sbi(struct super_block *sb)
 134 {
 135         struct ll_sb_info *sbi = ll_s2sbi(sb);
 136
 137         if (sbi->ll_cache) {
 138                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 139                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 140                 cl_cache_decref(sbi->ll_cache);
 141                 sbi->ll_cache = NULL;
 142         }
 143
 144         kfree(sbi);
 145 }
 146
 147 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 148                                     struct vfsmount *mnt)
 149 {
 150         struct inode *root = NULL;
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         struct obd_device *obd;
 153         struct obd_statfs *osfs = NULL;
 154         struct ptlrpc_request *request = NULL;
 155         struct obd_connect_data *data = NULL;
 156         struct obd_uuid *uuid;
 157         struct md_op_data *op_data;
 158         struct lustre_md lmd;
 159         u64 valid;
 160         int size, err, checksum;
 161
 162         obd = class_name2obd(md);
 163         if (!obd) {
 164                 CERROR("MD %s: not setup or attached\n", md);
 165                 return -EINVAL;
 166         }
 167
 168         data = kzalloc(sizeof(*data), GFP_NOFS);
 169         if (!data)
 170                 return -ENOMEM;
 171
 172         osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
 173         if (!osfs) {
 174                 kfree(data);
 175                 return -ENOMEM;
 176         }
 177
 178         /* indicate the features supported by this client */
 179         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 180                                   OBD_CONNECT_ATTRFID  |
 181                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 182                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 183                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 184                                   OBD_CONNECT_VBR       | OBD_CONNECT_FULL20  |
 185                                   OBD_CONNECT_64BITHASH |
 186                                   OBD_CONNECT_EINPROGRESS |
 187                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 188                                   OBD_CONNECT_LAYOUTLOCK |
 189                                   OBD_CONNECT_PINGLESS |
 190                                   OBD_CONNECT_MAX_EASIZE |
 191                                   OBD_CONNECT_FLOCK_DEAD |
 192                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 193                                   OBD_CONNECT_OPEN_BY_FID |
 194                                   OBD_CONNECT_DIR_STRIPE;
 195
 196         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 197                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 198
 199         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 200                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 201 #ifdef CONFIG_FS_POSIX_ACL
 202         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 203 #endif
 204
 205         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 206                 /* flag mdc connection as lightweight, only used for test
 207                  * purpose, use with care
 208                  */
 209                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 210
 211         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 212         data->ocd_version = LUSTRE_VERSION_CODE;
 213
 214         if (sb->s_flags & MS_RDONLY)
 215                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 216         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 217                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 218
 219         if (sbi->ll_flags & LL_SBI_FLOCK)
 220                 sbi->ll_fop = &ll_file_operations_flock;
 221         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 222                 sbi->ll_fop = &ll_file_operations;
 223         else
 224                 sbi->ll_fop = &ll_file_operations_noflock;
 225
 226         /* real client */
 227         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 228
 229         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 230
 231         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
 232                           data, NULL);
 233         if (err == -EBUSY) {
 234                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
 235                                    md);
 236                 goto out;
 237         } else if (err) {
 238                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 239                 goto out;
 240         }
 241
 242         sbi->ll_md_exp->exp_connect_data = *data;
 243
 244         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 245                            LUSTRE_SEQ_METADATA);
 246         if (err) {
 247                 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
 248                        sbi->ll_md_exp->exp_obd->obd_name, err);
 249                 goto out_md;
 250         }
 251
 252         /* For mount, we only need fs info from MDT0, and also in DNE, it
 253          * can make sure the client can be mounted as long as MDT0 is
 254          * available
 255          */
 256         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 257                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 258                          OBD_STATFS_FOR_MDT0);
 259         if (err)
 260                 goto out_md_fid;
 261
 262         /* This needs to be after statfs to ensure connect has finished.
 263          * Note that "data" does NOT contain the valid connect reply.
 264          * If connecting to a 1.8 server there will be no LMV device, so
 265          * we can access the MDC export directly and exp_connect_flags will
 266          * be non-zero, but if accessing an upgraded 2.1 server it will
 267          * have the correct flags filled in.
 268          * XXX: fill in the LMV exp_connect_flags from MDC(s).
 269          */
 270         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 271         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 272             valid != CLIENT_CONNECT_MDT_REQD) {
 273                 char *buf;
 274
 275                 buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
 276                 if (!buf) {
 277                         err = -ENOMEM;
 278                         goto out_md_fid;
 279                 }
 280                 obd_connect_flags2str(buf, PAGE_SIZE,
 281                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 282                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support feature(s) needed for correct operation of this client (%s). Please upgrade server or downgrade client.\n",
 283                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 284                 kfree(buf);
 285                 err = -EPROTO;
 286                 goto out_md_fid;
 287         }
 288
 289         size = sizeof(*data);
 290         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 291                            KEY_CONN_DATA,  &size, data, NULL);
 292         if (err) {
 293                 CERROR("%s: Get connect data failed: rc = %d\n",
 294                        sbi->ll_md_exp->exp_obd->obd_name, err);
 295                 goto out_md_fid;
 296         }
 297
 298         LASSERT(osfs->os_bsize);
 299         sb->s_blocksize = osfs->os_bsize;
 300         sb->s_blocksize_bits = log2(osfs->os_bsize);
 301         sb->s_magic = LL_SUPER_MAGIC;
 302         sb->s_maxbytes = MAX_LFS_FILESIZE;
 303         sbi->ll_namelen = osfs->os_namelen;
 304
 305         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 306             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 307                 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
 308                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 309         }
 310
 311         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 312                 sb->s_flags |= MS_POSIXACL;
 313                 sbi->ll_flags |= LL_SBI_ACL;
 314         } else {
 315                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 316                 sb->s_flags &= ~MS_POSIXACL;
 317                 sbi->ll_flags &= ~LL_SBI_ACL;
 318         }
 319
 320         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 321                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 322
 323         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 324                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_SHIFT;
 325         else
 326                 sbi->ll_md_brw_pages = 1;
 327
 328         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 329                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 330
 331         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 332                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 333                         LCONSOLE_INFO(
 334                                 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
 335                                 dt);
 336                 } else {
 337                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 338                         sbi->ll_xattr_cache_enabled = 1;
 339                 }
 340         }
 341
 342         obd = class_name2obd(dt);
 343         if (!obd) {
 344                 CERROR("DT %s: not setup or attached\n", dt);
 345                 err = -ENODEV;
 346                 goto out_md_fid;
 347         }
 348
 349         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 350                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 351                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 352                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 353                                   OBD_CONNECT_AT        | OBD_CONNECT_OSS_CAPA |
 354                                   OBD_CONNECT_VBR       | OBD_CONNECT_FULL20   |
 355                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 356                                   OBD_CONNECT_EINPROGRESS |
 357                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 358                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 359
 360         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 361                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 362
 363         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 364                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 365                  * disabled by default, because it can still be enabled on the
 366                  * fly via /sys. As a consequence, we still need to come to an
 367                  * agreement on the supported algorithms at connect time
 368                  */
 369                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 370
 371                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 372                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 373                 else
 374                         data->ocd_cksum_types = cksum_types_supported_client();
 375         }
 376
 377         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 378
 379         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
 380                data->ocd_connect_flags,
 381                data->ocd_version, data->ocd_grant);
 382
 383         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 384         obd->obd_upcall.onu_upcall = cl_ocd_update;
 385
 386         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 387
 388         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 389                           NULL);
 390         if (err == -EBUSY) {
 391                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing recovery, of which this client is not a part.  Please wait for recovery to complete, abort, or time out.\n",
 392                                    dt);
 393                 goto out_md;
 394         } else if (err) {
 395                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 396                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 397                 goto out_md;
 398         }
 399
 400         sbi->ll_dt_exp->exp_connect_data = *data;
 401
 402         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 403                            LUSTRE_SEQ_METADATA);
 404         if (err) {
 405                 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
 406                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 407                 goto out_dt;
 408         }
 409
 410         mutex_lock(&sbi->ll_lco.lco_lock);
 411         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 412         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 413         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 414         mutex_unlock(&sbi->ll_lco.lco_lock);
 415
 416         fid_zero(&sbi->ll_root_fid);
 417         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
 418         if (err) {
 419                 CERROR("cannot mds_connect: rc = %d\n", err);
 420                 goto out_lock_cn_cb;
 421         }
 422         if (!fid_is_sane(&sbi->ll_root_fid)) {
 423                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 424                        sbi->ll_md_exp->exp_obd->obd_name,
 425                        PFID(&sbi->ll_root_fid));
 426                 err = -EINVAL;
 427                 goto out_lock_cn_cb;
 428         }
 429         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 430
 431         sb->s_op = &lustre_super_operations;
 432         sb->s_xattr = ll_xattr_handlers;
 433 #if THREAD_SIZE >= 8192 /*b=17630*/
 434         sb->s_export_op = &lustre_export_operations;
 435 #endif
 436
 437         /* make root inode
 438          * XXX: move this to after cbd setup?
 439          */
 440         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 441         if (sbi->ll_flags & LL_SBI_ACL)
 442                 valid |= OBD_MD_FLACL;
 443
 444         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
 445         if (!op_data) {
 446                 err = -ENOMEM;
 447                 goto out_lock_cn_cb;
 448         }
 449
 450         op_data->op_fid1 = sbi->ll_root_fid;
 451         op_data->op_mode = 0;
 452         op_data->op_valid = valid;
 453
 454         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 455         kfree(op_data);
 456         if (err) {
 457                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 458                        sbi->ll_md_exp->exp_obd->obd_name, err);
 459                 goto out_lock_cn_cb;
 460         }
 461
 462         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 463                                sbi->ll_md_exp, &lmd);
 464         if (err) {
 465                 CERROR("failed to understand root inode md: rc = %d\n", err);
 466                 ptlrpc_req_finished(request);
 467                 goto out_lock_cn_cb;
 468         }
 469
 470         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 471         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 472                                             sbi->ll_flags & LL_SBI_32BIT_API),
 473                        &lmd);
 474         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 475         ptlrpc_req_finished(request);
 476
 477         if (IS_ERR(root)) {
 478                 if (lmd.lsm)
 479                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 480 #ifdef CONFIG_FS_POSIX_ACL
 481                 if (lmd.posix_acl) {
 482                         posix_acl_release(lmd.posix_acl);
 483                         lmd.posix_acl = NULL;
 484                 }
 485 #endif
 486                 err = -EBADF;
 487                 CERROR("lustre_lite: bad iget4 for root\n");
 488                 goto out_root;
 489         }
 490
 491         err = ll_close_thread_start(&sbi->ll_lcq);
 492         if (err) {
 493                 CERROR("cannot start close thread: rc %d\n", err);
 494                 goto out_root;
 495         }
 496
 497         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 498         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 499                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 500                                  NULL);
 501         cl_sb_init(sb);
 502
 503         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 504                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 505                                  sbi->ll_cache, NULL);
 506
 507         sb->s_root = d_make_root(root);
 508         if (!sb->s_root) {
 509                 CERROR("%s: can't make root dentry\n",
 510                        ll_get_fsname(sb, NULL, 0));
 511                 err = -ENOMEM;
 512                 goto out_lock_cn_cb;
 513         }
 514
 515         sbi->ll_sdev_orig = sb->s_dev;
 516
 517         /* We set sb->s_dev equal on all lustre clients in order to support
 518          * NFS export clustering.  NFSD requires that the FSID be the same
 519          * on all clients.
 520          */
 521         /* s_dev is also used in lt_compare() to compare two fs, but that is
 522          * only a node-local comparison.
 523          */
 524         uuid = obd_get_uuid(sbi->ll_md_exp);
 525         if (uuid) {
 526                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 527                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 528         }
 529
 530         kfree(data);
 531         kfree(osfs);
 532
 533         if (llite_root) {
 534                 err = ldebugfs_register_mountpoint(llite_root, sb, dt, md);
 535                 if (err < 0) {
 536                         CERROR("%s: could not register mount in debugfs: "
 537                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 538                         err = 0;
 539                 }
 540         }
 541
 542         return err;
 543 out_root:
 544         iput(root);
 545 out_lock_cn_cb:
 546         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 547 out_dt:
 548         obd_disconnect(sbi->ll_dt_exp);
 549         sbi->ll_dt_exp = NULL;
 550 out_md_fid:
 551         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 552 out_md:
 553         obd_disconnect(sbi->ll_md_exp);
 554         sbi->ll_md_exp = NULL;
 555 out:
 556         kfree(data);
 557         kfree(osfs);
 558         return err;
 559 }
 560
 561 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 562 {
 563         int size, rc;
 564
 565         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 566         size = sizeof(int);
 567         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 568                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 569         if (rc)
 570                 CERROR("Get max mdsize error rc %d\n", rc);
 571
 572         return rc;
 573 }
 574
 575 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 576 {
 577         int size, rc;
 578
 579         size = sizeof(int);
 580         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 581                           KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 582         if (rc)
 583                 CERROR("Get default mdsize error rc %d\n", rc);
 584
 585         return rc;
 586 }
 587
 588 static void client_common_put_super(struct super_block *sb)
 589 {
 590         struct ll_sb_info *sbi = ll_s2sbi(sb);
 591
 592         ll_close_thread_shutdown(sbi->ll_lcq);
 593
 594         cl_sb_fini(sb);
 595
 596         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 597         obd_disconnect(sbi->ll_dt_exp);
 598         sbi->ll_dt_exp = NULL;
 599
 600         ldebugfs_unregister_mountpoint(sbi);
 601
 602         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 603         obd_disconnect(sbi->ll_md_exp);
 604         sbi->ll_md_exp = NULL;
 605 }
 606
 607 void ll_kill_super(struct super_block *sb)
 608 {
 609         struct ll_sb_info *sbi;
 610
 611         /* not init sb ?*/
 612         if (!(sb->s_flags & MS_ACTIVE))
 613                 return;
 614
 615         sbi = ll_s2sbi(sb);
 616         /* we need to restore s_dev from changed for clustered NFS before
 617          * put_super because new kernels have cached s_dev and change sb->s_dev
 618          * in put_super not affected real removing devices
 619          */
 620         if (sbi) {
 621                 sb->s_dev = sbi->ll_sdev_orig;
 622                 sbi->ll_umounting = 1;
 623         }
 624 }
 625
 626 static inline int ll_set_opt(const char *opt, char *data, int fl)
 627 {
 628         if (strncmp(opt, data, strlen(opt)) != 0)
 629                 return 0;
 630         else
 631                 return fl;
 632 }
 633
 634 /* non-client-specific mount options are parsed in lmd_parse */
 635 static int ll_options(char *options, int *flags)
 636 {
 637         int tmp;
 638         char *s1 = options, *s2;
 639
 640         if (!options)
 641                 return 0;
 642
 643         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 644
 645         while (*s1) {
 646                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 647                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 648                 if (tmp) {
 649                         *flags |= tmp;
 650                         goto next;
 651                 }
 652                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 653                 if (tmp) {
 654                         *flags |= tmp;
 655                         goto next;
 656                 }
 657                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 658                 if (tmp) {
 659                         *flags |= tmp;
 660                         goto next;
 661                 }
 662                 tmp = ll_set_opt("noflock", s1,
 663                                  LL_SBI_FLOCK | LL_SBI_LOCALFLOCK);
 664                 if (tmp) {
 665                         *flags &= ~tmp;
 666                         goto next;
 667                 }
 668                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 669                 if (tmp) {
 670                         *flags |= tmp;
 671                         goto next;
 672                 }
 673                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 674                 if (tmp) {
 675                         *flags &= ~tmp;
 676                         goto next;
 677                 }
 678                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 679                 if (tmp) {
 680                         *flags |= tmp;
 681                         goto next;
 682                 }
 683                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 684                 if (tmp) {
 685                         *flags &= ~tmp;
 686                         goto next;
 687                 }
 688
 689                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 690                 if (tmp) {
 691                         *flags |= tmp;
 692                         goto next;
 693                 }
 694                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 695                 if (tmp) {
 696                         *flags &= ~tmp;
 697                         goto next;
 698                 }
 699                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 700                 if (tmp) {
 701                         *flags |= tmp;
 702                         goto next;
 703                 }
 704                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 705                 if (tmp) {
 706                         *flags &= ~tmp;
 707                         goto next;
 708                 }
 709                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 710                 if (tmp) {
 711                         *flags |= tmp;
 712                         goto next;
 713                 }
 714                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 715                 if (tmp) {
 716                         *flags &= ~tmp;
 717                         goto next;
 718                 }
 719                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 720                 if (tmp) {
 721                         *flags |= tmp;
 722                         goto next;
 723                 }
 724                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 725                 if (tmp) {
 726                         *flags |= tmp;
 727                         goto next;
 728                 }
 729                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 730                 if (tmp) {
 731                         *flags |= tmp;
 732                         goto next;
 733                 }
 734                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 735                 if (tmp) {
 736                         *flags &= ~tmp;
 737                         goto next;
 738                 }
 739                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 740                                    s1);
 741                 return -EINVAL;
 742
 743 next:
 744                 /* Find next opt */
 745                 s2 = strchr(s1, ',');
 746                 if (!s2)
 747                         break;
 748                 s1 = s2 + 1;
 749         }
 750         return 0;
 751 }
 752
 753 void ll_lli_init(struct ll_inode_info *lli)
 754 {
 755         lli->lli_inode_magic = LLI_INODE_MAGIC;
 756         lli->lli_flags = 0;
 757         lli->lli_ioepoch = 0;
 758         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 759         spin_lock_init(&lli->lli_lock);
 760         lli->lli_posix_acl = NULL;
 761         /* Do not set lli_fid, it has been initialized already. */
 762         fid_zero(&lli->lli_pfid);
 763         INIT_LIST_HEAD(&lli->lli_close_list);
 764         lli->lli_pending_och = NULL;
 765         lli->lli_mds_read_och = NULL;
 766         lli->lli_mds_write_och = NULL;
 767         lli->lli_mds_exec_och = NULL;
 768         lli->lli_open_fd_read_count = 0;
 769         lli->lli_open_fd_write_count = 0;
 770         lli->lli_open_fd_exec_count = 0;
 771         mutex_init(&lli->lli_och_mutex);
 772         spin_lock_init(&lli->lli_agl_lock);
 773         lli->lli_has_smd = false;
 774         spin_lock_init(&lli->lli_layout_lock);
 775         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 776         lli->lli_clob = NULL;
 777
 778         init_rwsem(&lli->lli_xattrs_list_rwsem);
 779         mutex_init(&lli->lli_xattrs_enq_lock);
 780
 781         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 782         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 783                 mutex_init(&lli->lli_readdir_mutex);
 784                 lli->lli_opendir_key = NULL;
 785                 lli->lli_sai = NULL;
 786                 spin_lock_init(&lli->lli_sa_lock);
 787                 lli->lli_opendir_pid = 0;
 788         } else {
 789                 mutex_init(&lli->lli_size_mutex);
 790                 lli->lli_symlink_name = NULL;
 791                 init_rwsem(&lli->lli_trunc_sem);
 792                 mutex_init(&lli->lli_write_mutex);
 793                 init_rwsem(&lli->lli_glimpse_sem);
 794                 lli->lli_glimpse_time = 0;
 795                 INIT_LIST_HEAD(&lli->lli_agl_list);
 796                 lli->lli_agl_index = 0;
 797                 lli->lli_async_rc = 0;
 798         }
 799         mutex_init(&lli->lli_layout_mutex);
 800 }
 801
 802 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 803 {
 804         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 805
 806         bdi->name = "lustre";
 807         return bdi_register(bdi, NULL, "lustre-%d",
 808                             atomic_inc_return(&ll_bdi_num));
 809 }
 810
 811 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 812 {
 813         struct lustre_profile *lprof = NULL;
 814         struct lustre_sb_info *lsi = s2lsi(sb);
 815         struct ll_sb_info *sbi;
 816         char  *dt = NULL, *md = NULL;
 817         char  *profilenm = get_profile_name(sb);
 818         struct config_llog_instance *cfg;
 819         int    err;
 820
 821         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 822
 823         cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
 824         if (!cfg)
 825                 return -ENOMEM;
 826
 827         try_module_get(THIS_MODULE);
 828
 829         /* client additional sb info */
 830         sbi = ll_init_sbi(sb);
 831         lsi->lsi_llsbi = sbi;
 832         if (!sbi) {
 833                 module_put(THIS_MODULE);
 834                 kfree(cfg);
 835                 return -ENOMEM;
 836         }
 837
 838         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 839         if (err)
 840                 goto out_free;
 841
 842         err = bdi_init(&lsi->lsi_bdi);
 843         if (err)
 844                 goto out_free;
 845         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 846         lsi->lsi_bdi.capabilities = 0;
 847         err = ll_bdi_register(&lsi->lsi_bdi);
 848         if (err)
 849                 goto out_free;
 850
 851         sb->s_bdi = &lsi->lsi_bdi;
 852         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 853         sb->s_d_op = &ll_d_ops;
 854
 855         /* Generate a string unique to this super, in case some joker tries
 856          * to mount the same fs at two mount points.
 857          * Use the address of the super itself.
 858          */
 859         cfg->cfg_instance = sb;
 860         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 861         cfg->cfg_callback = class_config_llog_handler;
 862         /* set up client obds */
 863         err = lustre_process_log(sb, profilenm, cfg);
 864         if (err < 0)
 865                 goto out_free;
 866
 867         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 868         lprof = class_get_profile(profilenm);
 869         if (!lprof) {
 870                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS.  Does that filesystem exist?\n",
 871                                    profilenm);
 872                 err = -EINVAL;
 873                 goto out_free;
 874         }
 875         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 876                lprof->lp_md, lprof->lp_dt);
 877
 878         dt = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 879         if (!dt) {
 880                 err = -ENOMEM;
 881                 goto out_free;
 882         }
 883
 884         md = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_md, cfg->cfg_instance);
 885         if (!md) {
 886                 err = -ENOMEM;
 887                 goto out_free;
 888         }
 889
 890         /* connections, registrations, sb setup */
 891         err = client_common_fill_super(sb, md, dt, mnt);
 892
 893 out_free:
 894         kfree(md);
 895         kfree(dt);
 896         if (err)
 897                 ll_put_super(sb);
 898         else if (sbi->ll_flags & LL_SBI_VERBOSE)
 899                 LCONSOLE_WARN("Mounted %s\n", profilenm);
 900
 901         kfree(cfg);
 902         return err;
 903 } /* ll_fill_super */
 904
 905 void ll_put_super(struct super_block *sb)
 906 {
 907         struct config_llog_instance cfg, params_cfg;
 908         struct obd_device *obd;
 909         struct lustre_sb_info *lsi = s2lsi(sb);
 910         struct ll_sb_info *sbi = ll_s2sbi(sb);
 911         char *profilenm = get_profile_name(sb);
 912         int ccc_count, next, force = 1, rc = 0;
 913
 914         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 915
 916         cfg.cfg_instance = sb;
 917         lustre_end_log(sb, profilenm, &cfg);
 918
 919         params_cfg.cfg_instance = sb;
 920         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
 921
 922         if (sbi->ll_md_exp) {
 923                 obd = class_exp2obd(sbi->ll_md_exp);
 924                 if (obd)
 925                         force = obd->obd_force;
 926         }
 927
 928         /* Wait for unstable pages to be committed to stable storage */
 929         if (!force) {
 930                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
 931
 932                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
 933                                   !atomic_read(&sbi->ll_cache->ccc_unstable_nr),
 934                                   &lwi);
 935         }
 936
 937         ccc_count = atomic_read(&sbi->ll_cache->ccc_unstable_nr);
 938         if (!force && rc != -EINTR)
 939                 LASSERTF(!ccc_count, "count: %i\n", ccc_count);
 940
 941         /* We need to set force before the lov_disconnect in
 942          * lustre_common_put_super, since l_d cleans up osc's as well.
 943          */
 944         if (force) {
 945                 next = 0;
 946                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 947                                                      &next)) != NULL) {
 948                         obd->obd_force = force;
 949                 }
 950         }
 951
 952         if (sbi->ll_lcq) {
 953                 /* Only if client_common_fill_super succeeded */
 954                 client_common_put_super(sb);
 955         }
 956
 957         next = 0;
 958         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
 959                 class_manual_cleanup(obd);
 960
 961         if (sbi->ll_flags & LL_SBI_VERBOSE)
 962                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
 963
 964         if (profilenm)
 965                 class_del_profile(profilenm);
 966
 967         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
 968                 bdi_destroy(&lsi->lsi_bdi);
 969                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
 970         }
 971
 972         ll_free_sbi(sb);
 973         lsi->lsi_llsbi = NULL;
 974
 975         lustre_common_put_super(sb);
 976
 977         cl_env_cache_purge(~0);
 978
 979         module_put(THIS_MODULE);
 980 } /* client_put_super */
 981
 982 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
 983 {
 984         struct inode *inode = NULL;
 985
 986         /* NOTE: we depend on atomic igrab() -bzzz */
 987         lock_res_and_lock(lock);
 988         if (lock->l_resource->lr_lvb_inode) {
 989                 struct ll_inode_info *lli;
 990
 991                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
 992                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 993                         inode = igrab(lock->l_resource->lr_lvb_inode);
 994                 } else {
 995                         inode = lock->l_resource->lr_lvb_inode;
 996                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
 997                                          D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
 998                                          lock->l_resource->lr_lvb_inode,
 999                                          lli->lli_inode_magic);
1000                         inode = NULL;
1001                 }
1002         }
1003         unlock_res_and_lock(lock);
1004         return inode;
1005 }
1006
1007 static void ll_dir_clear_lsm_md(struct inode *inode)
1008 {
1009         struct ll_inode_info *lli = ll_i2info(inode);
1010
1011         LASSERT(S_ISDIR(inode->i_mode));
1012
1013         if (lli->lli_lsm_md) {
1014                 lmv_free_memmd(lli->lli_lsm_md);
1015                 lli->lli_lsm_md = NULL;
1016         }
1017 }
1018
1019 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1020                                       const struct lu_fid *fid,
1021                                       struct lustre_md *md)
1022 {
1023         struct ll_sb_info *sbi = ll_s2sbi(sb);
1024         struct mdt_body *body = md->body;
1025         struct inode *inode;
1026         ino_t ino;
1027
1028         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1029         inode = iget_locked(sb, ino);
1030         if (!inode) {
1031                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1032                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1033                 return ERR_PTR(-ENOENT);
1034         }
1035
1036         if (inode->i_state & I_NEW) {
1037                 struct ll_inode_info *lli = ll_i2info(inode);
1038                 struct lmv_stripe_md *lsm = md->lmv;
1039
1040                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1041                                 (body->mbo_mode & S_IFMT);
1042                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1043                          PFID(fid));
1044
1045                 LTIME_S(inode->i_mtime) = 0;
1046                 LTIME_S(inode->i_atime) = 0;
1047                 LTIME_S(inode->i_ctime) = 0;
1048                 inode->i_rdev = 0;
1049
1050                 inode->i_op = &ll_dir_inode_operations;
1051                 inode->i_fop = &ll_dir_operations;
1052                 lli->lli_fid = *fid;
1053                 ll_lli_init(lli);
1054
1055                 LASSERT(lsm);
1056                 /* master object FID */
1057                 lli->lli_pfid = body->mbo_fid1;
1058                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1059                        lli, PFID(fid), PFID(&lli->lli_pfid));
1060                 unlock_new_inode(inode);
1061         }
1062
1063         return inode;
1064 }
1065
1066 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1067 {
1068         struct lmv_stripe_md *lsm = md->lmv;
1069         struct lu_fid *fid;
1070         int i;
1071
1072         LASSERT(lsm);
1073         /*
1074          * XXX sigh, this lsm_root initialization should be in
1075          * LMV layer, but it needs ll_iget right now, so we
1076          * put this here right now.
1077          */
1078         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1079                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1080                 LASSERT(!lsm->lsm_md_oinfo[i].lmo_root);
1081                 /* Unfortunately ll_iget will call ll_update_inode,
1082                  * where the initialization of slave inode is slightly
1083                  * different, so it reset lsm_md to NULL to avoid
1084                  * initializing lsm for slave inode.
1085                  */
1086                 /* For migrating inode, master stripe and master object will
1087                  * be same, so we only need assign this inode
1088                  */
1089                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && !i)
1090                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1091                 else
1092                         lsm->lsm_md_oinfo[i].lmo_root =
1093                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1094                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1095                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1096
1097                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1098                         return rc;
1099                 }
1100         }
1101
1102         /*
1103          * Here is where the lsm is being initialized(fill lmo_info) after
1104          * client retrieve MD stripe information from MDT.
1105          */
1106         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1107                                 ll_md_blocking_ast);
1108 }
1109
1110 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1111                                 const struct lmv_stripe_md *lsm_md2)
1112 {
1113         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1114                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1115                lsm_md1->lsm_md_master_mdt_index ==
1116                         lsm_md2->lsm_md_master_mdt_index &&
1117                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1118                lsm_md1->lsm_md_layout_version ==
1119                         lsm_md2->lsm_md_layout_version &&
1120                !strcmp(lsm_md1->lsm_md_pool_name,
1121                        lsm_md2->lsm_md_pool_name);
1122 }
1123
1124 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1125 {
1126         struct ll_inode_info *lli = ll_i2info(inode);
1127         struct lmv_stripe_md *lsm = md->lmv;
1128         int rc;
1129
1130         LASSERT(S_ISDIR(inode->i_mode));
1131         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1132                PFID(ll_inode2fid(inode)));
1133
1134         /* no striped information from request. */
1135         if (!lsm) {
1136                 if (!lli->lli_lsm_md) {
1137                         return 0;
1138                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1139                            LMV_HASH_FLAG_MIGRATION) {
1140                         /*
1141                          * migration is done, the temporay MIGRATE layout has
1142                          * been removed
1143                          */
1144                         CDEBUG(D_INODE, DFID" finish migration.\n",
1145                                PFID(ll_inode2fid(inode)));
1146                         lmv_free_memmd(lli->lli_lsm_md);
1147                         lli->lli_lsm_md = NULL;
1148                         return 0;
1149                 } else {
1150                         /*
1151                          * The lustre_md from req does not include stripeEA,
1152                          * see ll_md_setattr
1153                          */
1154                         return 0;
1155                 }
1156         }
1157
1158         /* set the directory layout */
1159         if (!lli->lli_lsm_md) {
1160                 rc = ll_init_lsm_md(inode, md);
1161                 if (rc)
1162                         return rc;
1163
1164                 lli->lli_lsm_md = lsm;
1165                 /*
1166                  * set lsm_md to NULL, so the following free lustre_md
1167                  * will not free this lsm
1168                  */
1169                 md->lmv = NULL;
1170                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1171                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1172                 return 0;
1173         }
1174
1175         /* Compare the old and new stripe information */
1176         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1177                 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1178                 int idx;
1179
1180                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p) magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1181                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1182                        inode, lsm, old_lsm,
1183                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1184                        lsm->lsm_md_stripe_count,
1185                        old_lsm->lsm_md_stripe_count,
1186                        lsm->lsm_md_master_mdt_index,
1187                        old_lsm->lsm_md_master_mdt_index,
1188                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1189                        lsm->lsm_md_layout_version,
1190                        old_lsm->lsm_md_layout_version,
1191                        lsm->lsm_md_pool_name,
1192                        old_lsm->lsm_md_pool_name);
1193
1194                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1195                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1196                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1197                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1198                 }
1199
1200                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1201                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1202                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1203                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1204                 }
1205
1206                 return -EIO;
1207         }
1208
1209         return 0;
1210 }
1211
1212 void ll_clear_inode(struct inode *inode)
1213 {
1214         struct ll_inode_info *lli = ll_i2info(inode);
1215         struct ll_sb_info *sbi = ll_i2sbi(inode);
1216
1217         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1218                PFID(ll_inode2fid(inode)), inode);
1219
1220         if (S_ISDIR(inode->i_mode)) {
1221                 /* these should have been cleared in ll_file_release */
1222                 LASSERT(!lli->lli_opendir_key);
1223                 LASSERT(!lli->lli_sai);
1224                 LASSERT(lli->lli_opendir_pid == 0);
1225         }
1226
1227         spin_lock(&lli->lli_lock);
1228         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1229         spin_unlock(&lli->lli_lock);
1230         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1231
1232         LASSERT(!lli->lli_open_fd_write_count);
1233         LASSERT(!lli->lli_open_fd_read_count);
1234         LASSERT(!lli->lli_open_fd_exec_count);
1235
1236         if (lli->lli_mds_write_och)
1237                 ll_md_real_close(inode, FMODE_WRITE);
1238         if (lli->lli_mds_exec_och)
1239                 ll_md_real_close(inode, FMODE_EXEC);
1240         if (lli->lli_mds_read_och)
1241                 ll_md_real_close(inode, FMODE_READ);
1242
1243         if (S_ISLNK(inode->i_mode)) {
1244                 kfree(lli->lli_symlink_name);
1245                 lli->lli_symlink_name = NULL;
1246         }
1247
1248         ll_xattr_cache_destroy(inode);
1249
1250 #ifdef CONFIG_FS_POSIX_ACL
1251         if (lli->lli_posix_acl) {
1252                 posix_acl_release(lli->lli_posix_acl);
1253                 lli->lli_posix_acl = NULL;
1254         }
1255 #endif
1256         lli->lli_inode_magic = LLI_INODE_DEAD;
1257
1258         if (S_ISDIR(inode->i_mode))
1259                 ll_dir_clear_lsm_md(inode);
1260         if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1261                 LASSERT(list_empty(&lli->lli_agl_list));
1262
1263         /*
1264          * XXX This has to be done before lsm is freed below, because
1265          * cl_object still uses inode lsm.
1266          */
1267         cl_inode_fini(inode);
1268         lli->lli_has_smd = false;
1269 }
1270
1271 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
1272
1273 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1274                          struct md_open_data **mod)
1275 {
1276         struct lustre_md md;
1277         struct inode *inode = d_inode(dentry);
1278         struct ll_sb_info *sbi = ll_i2sbi(inode);
1279         struct ptlrpc_request *request = NULL;
1280         int rc, ia_valid;
1281
1282         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1283                                      LUSTRE_OPC_ANY, NULL);
1284         if (IS_ERR(op_data))
1285                 return PTR_ERR(op_data);
1286
1287         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1288                         &request, mod);
1289         if (rc) {
1290                 ptlrpc_req_finished(request);
1291                 if (rc == -ENOENT) {
1292                         clear_nlink(inode);
1293                         /* Unlinked special device node? Or just a race?
1294                          * Pretend we did everything.
1295                          */
1296                         if (!S_ISREG(inode->i_mode) &&
1297                             !S_ISDIR(inode->i_mode)) {
1298                                 ia_valid = op_data->op_attr.ia_valid;
1299                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1300                                 rc = simple_setattr(dentry, &op_data->op_attr);
1301                                 op_data->op_attr.ia_valid = ia_valid;
1302                         }
1303                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1304                         CERROR("md_setattr fails: rc = %d\n", rc);
1305                 }
1306                 return rc;
1307         }
1308
1309         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1310                               sbi->ll_md_exp, &md);
1311         if (rc) {
1312                 ptlrpc_req_finished(request);
1313                 return rc;
1314         }
1315
1316         ia_valid = op_data->op_attr.ia_valid;
1317         /* inode size will be in cl_setattr_ost, can't do it now since dirty
1318          * cache is not cleared yet.
1319          */
1320         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1321         rc = simple_setattr(dentry, &op_data->op_attr);
1322         op_data->op_attr.ia_valid = ia_valid;
1323
1324         /* Extract epoch data if obtained. */
1325         op_data->op_handle = md.body->mbo_handle;
1326         op_data->op_ioepoch = md.body->mbo_ioepoch;
1327
1328         rc = ll_update_inode(inode, &md);
1329         ptlrpc_req_finished(request);
1330
1331         return rc;
1332 }
1333
1334 /* Close IO epoch and send Size-on-MDS attribute update. */
1335 static int ll_setattr_done_writing(struct inode *inode,
1336                                    struct md_op_data *op_data,
1337                                    struct md_open_data *mod)
1338 {
1339         struct ll_inode_info *lli = ll_i2info(inode);
1340         int rc = 0;
1341
1342         if (!S_ISREG(inode->i_mode))
1343                 return 0;
1344
1345         CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1346                op_data->op_ioepoch, PFID(&lli->lli_fid));
1347
1348         op_data->op_flags = MF_EPOCH_CLOSE;
1349         ll_done_writing_attr(inode, op_data);
1350         ll_pack_inode2opdata(inode, op_data, NULL);
1351
1352         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1353         if (rc == -EAGAIN)
1354                 /* MDS has instructed us to obtain Size-on-MDS attribute
1355                  * from OSTs and send setattr to back to MDS.
1356                  */
1357                 rc = ll_som_update(inode, op_data);
1358         else if (rc) {
1359                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1360                       ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1361                       PFID(ll_inode2fid(inode)), rc);
1362         }
1363         return rc;
1364 }
1365
1366 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1367  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1368  * keep these values until such a time that objects are allocated for it.
1369  * We do the MDS operations first, as it is checking permissions for us.
1370  * We don't to the MDS RPC if there is nothing that we want to store there,
1371  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1372  * going to do an RPC anyways.
1373  *
1374  * If we are doing a truncate, we will send the mtime and ctime updates
1375  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1376  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1377  * at the same time.
1378  *
1379  * In case of HSMimport, we only set attr on MDS.
1380  */
1381 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1382 {
1383         struct inode *inode = d_inode(dentry);
1384         struct ll_inode_info *lli = ll_i2info(inode);
1385         struct md_op_data *op_data = NULL;
1386         struct md_open_data *mod = NULL;
1387         bool file_is_released = false;
1388         int rc = 0, rc1 = 0;
1389
1390         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, valid %x, hsm_import %d\n",
1391                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid), inode,
1392                i_size_read(inode), attr->ia_size, attr->ia_valid, hsm_import);
1393
1394         if (attr->ia_valid & ATTR_SIZE) {
1395                 /* Check new size against VFS/VM file size limit and rlimit */
1396                 rc = inode_newsize_ok(inode, attr->ia_size);
1397                 if (rc)
1398                         return rc;
1399
1400                 /* The maximum Lustre file size is variable, based on the
1401                  * OST maximum object size and number of stripes.  This
1402                  * needs another check in addition to the VFS check above.
1403                  */
1404                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1405                         CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1406                                PFID(&lli->lli_fid), attr->ia_size,
1407                                ll_file_maxbytes(inode));
1408                         return -EFBIG;
1409                 }
1410
1411                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1412         }
1413
1414         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1415         if (attr->ia_valid & TIMES_SET_FLAGS) {
1416                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1417                     !capable(CFS_CAP_FOWNER))
1418                         return -EPERM;
1419         }
1420
1421         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1422         if (attr->ia_valid & ATTR_CTIME) {
1423                 attr->ia_ctime = CURRENT_TIME;
1424                 attr->ia_valid |= ATTR_CTIME_SET;
1425         }
1426         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1427             (attr->ia_valid & ATTR_ATIME)) {
1428                 attr->ia_atime = CURRENT_TIME;
1429                 attr->ia_valid |= ATTR_ATIME_SET;
1430         }
1431         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1432             (attr->ia_valid & ATTR_MTIME)) {
1433                 attr->ia_mtime = CURRENT_TIME;
1434                 attr->ia_valid |= ATTR_MTIME_SET;
1435         }
1436
1437         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1438                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1439                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1440                        (s64)ktime_get_real_seconds());
1441
1442         /* We always do an MDS RPC, even if we're only changing the size;
1443          * only the MDS knows whether truncate() should fail with -ETXTBUSY
1444          */
1445
1446         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1447         if (!op_data)
1448                 return -ENOMEM;
1449
1450         if (!S_ISDIR(inode->i_mode))
1451                 inode_unlock(inode);
1452
1453         /* truncate on a released file must failed with -ENODATA,
1454          * so size must not be set on MDS for released file
1455          * but other attributes must be set
1456          */
1457         if (S_ISREG(inode->i_mode)) {
1458                 struct lov_stripe_md *lsm;
1459                 __u32 gen;
1460
1461                 ll_layout_refresh(inode, &gen);
1462                 lsm = ccc_inode_lsm_get(inode);
1463                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1464                         file_is_released = true;
1465                 ccc_inode_lsm_put(inode, lsm);
1466
1467                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1468                         if (file_is_released) {
1469                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1470                                 if (rc < 0)
1471                                         goto out;
1472
1473                                 file_is_released = false;
1474                                 ll_layout_refresh(inode, &gen);
1475                         }
1476
1477                         /*
1478                          * If we are changing file size, file content is
1479                          * modified, flag it.
1480                          */
1481                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1482                         spin_lock(&lli->lli_lock);
1483                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1484                         spin_unlock(&lli->lli_lock);
1485                         op_data->op_bias |= MDS_DATA_MODIFIED;
1486                 }
1487         }
1488
1489         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1490
1491         /* Open epoch for truncate. */
1492         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1493             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1494                 op_data->op_flags = MF_EPOCH_OPEN;
1495
1496         rc = ll_md_setattr(dentry, op_data, &mod);
1497         if (rc)
1498                 goto out;
1499
1500         /* RPC to MDT is sent, cancel data modification flag */
1501         if (op_data->op_bias & MDS_DATA_MODIFIED) {
1502                 spin_lock(&lli->lli_lock);
1503                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1504                 spin_unlock(&lli->lli_lock);
1505         }
1506
1507         ll_ioepoch_open(lli, op_data->op_ioepoch);
1508         if (!S_ISREG(inode->i_mode) || file_is_released) {
1509                 rc = 0;
1510                 goto out;
1511         }
1512
1513         if (attr->ia_valid & (ATTR_SIZE |
1514                               ATTR_ATIME | ATTR_ATIME_SET |
1515                               ATTR_MTIME | ATTR_MTIME_SET)) {
1516                 /* For truncate and utimes sending attributes to OSTs, setting
1517                  * mtime/atime to the past will be performed under PW [0:EOF]
1518                  * extent lock (new_size:EOF for truncate).  It may seem
1519                  * excessive to send mtime/atime updates to OSTs when not
1520                  * setting times to past, but it is necessary due to possible
1521                  * time de-synchronization between MDT inode and OST objects
1522                  */
1523                 if (attr->ia_valid & ATTR_SIZE)
1524                         down_write(&lli->lli_trunc_sem);
1525                 rc = cl_setattr_ost(inode, attr);
1526                 if (attr->ia_valid & ATTR_SIZE)
1527                         up_write(&lli->lli_trunc_sem);
1528         }
1529 out:
1530         if (op_data->op_ioepoch) {
1531                 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1532                 if (!rc)
1533                         rc = rc1;
1534         }
1535         ll_finish_md_op_data(op_data);
1536
1537         if (!S_ISDIR(inode->i_mode)) {
1538                 inode_lock(inode);
1539                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1540                         inode_dio_wait(inode);
1541         }
1542
1543         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1544                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1545
1546         return rc;
1547 }
1548
1549 int ll_setattr(struct dentry *de, struct iattr *attr)
1550 {
1551         int mode = d_inode(de)->i_mode;
1552
1553         if ((attr->ia_valid & (ATTR_CTIME | ATTR_SIZE | ATTR_MODE)) ==
1554                               (ATTR_CTIME | ATTR_SIZE | ATTR_MODE))
1555                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1556
1557         if (((attr->ia_valid & (ATTR_MODE | ATTR_FORCE | ATTR_SIZE)) ==
1558                                (ATTR_SIZE | ATTR_MODE)) &&
1559             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1560              (((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) &&
1561               !(attr->ia_mode & S_ISGID))))
1562                 attr->ia_valid |= ATTR_FORCE;
1563
1564         if ((attr->ia_valid & ATTR_MODE) &&
1565             (mode & S_ISUID) &&
1566             !(attr->ia_mode & S_ISUID) &&
1567             !(attr->ia_valid & ATTR_KILL_SUID))
1568                 attr->ia_valid |= ATTR_KILL_SUID;
1569
1570         if ((attr->ia_valid & ATTR_MODE) &&
1571             ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) &&
1572             !(attr->ia_mode & S_ISGID) &&
1573             !(attr->ia_valid & ATTR_KILL_SGID))
1574                 attr->ia_valid |= ATTR_KILL_SGID;
1575
1576         return ll_setattr_raw(de, attr, false);
1577 }
1578
1579 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1580                        __u64 max_age, __u32 flags)
1581 {
1582         struct ll_sb_info *sbi = ll_s2sbi(sb);
1583         struct obd_statfs obd_osfs;
1584         int rc;
1585
1586         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1587         if (rc) {
1588                 CERROR("md_statfs fails: rc = %d\n", rc);
1589                 return rc;
1590         }
1591
1592         osfs->os_type = sb->s_magic;
1593
1594         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1595                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1596                osfs->os_files);
1597
1598         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1599                 flags |= OBD_STATFS_NODELAY;
1600
1601         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1602         if (rc) {
1603                 CERROR("obd_statfs fails: rc = %d\n", rc);
1604                 return rc;
1605         }
1606
1607         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1608                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1609                obd_osfs.os_files);
1610
1611         osfs->os_bsize = obd_osfs.os_bsize;
1612         osfs->os_blocks = obd_osfs.os_blocks;
1613         osfs->os_bfree = obd_osfs.os_bfree;
1614         osfs->os_bavail = obd_osfs.os_bavail;
1615
1616         /* If we don't have as many objects free on the OST as inodes
1617          * on the MDS, we reduce the total number of inodes to
1618          * compensate, so that the "inodes in use" number is correct.
1619          */
1620         if (obd_osfs.os_ffree < osfs->os_ffree) {
1621                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1622                         obd_osfs.os_ffree;
1623                 osfs->os_ffree = obd_osfs.os_ffree;
1624         }
1625
1626         return rc;
1627 }
1628
1629 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1630 {
1631         struct super_block *sb = de->d_sb;
1632         struct obd_statfs osfs;
1633         int rc;
1634
1635         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1636         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1637
1638         /* Some amount of caching on the client is allowed */
1639         rc = ll_statfs_internal(sb, &osfs,
1640                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1641                                 0);
1642         if (rc)
1643                 return rc;
1644
1645         statfs_unpack(sfs, &osfs);
1646
1647         /* We need to downshift for all 32-bit kernels, because we can't
1648          * tell if the kernel is being called via sys_statfs64() or not.
1649          * Stop before overflowing f_bsize - in which case it is better
1650          * to just risk EOVERFLOW if caller is using old sys_statfs().
1651          */
1652         if (sizeof(long) < 8) {
1653                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1654                         sfs->f_bsize <<= 1;
1655
1656                         osfs.os_blocks >>= 1;
1657                         osfs.os_bfree >>= 1;
1658                         osfs.os_bavail >>= 1;
1659                 }
1660         }
1661
1662         sfs->f_blocks = osfs.os_blocks;
1663         sfs->f_bfree = osfs.os_bfree;
1664         sfs->f_bavail = osfs.os_bavail;
1665         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1666         return 0;
1667 }
1668
1669 void ll_inode_size_lock(struct inode *inode)
1670 {
1671         struct ll_inode_info *lli;
1672
1673         LASSERT(!S_ISDIR(inode->i_mode));
1674
1675         lli = ll_i2info(inode);
1676         mutex_lock(&lli->lli_size_mutex);
1677 }
1678
1679 void ll_inode_size_unlock(struct inode *inode)
1680 {
1681         struct ll_inode_info *lli;
1682
1683         lli = ll_i2info(inode);
1684         mutex_unlock(&lli->lli_size_mutex);
1685 }
1686
1687 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1688 {
1689         struct ll_inode_info *lli = ll_i2info(inode);
1690         struct mdt_body *body = md->body;
1691         struct lov_stripe_md *lsm = md->lsm;
1692         struct ll_sb_info *sbi = ll_i2sbi(inode);
1693
1694         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1695         if (lsm) {
1696                 if (!lli->lli_has_smd &&
1697                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1698                         cl_file_inode_init(inode, md);
1699
1700                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1701                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1702                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1703         }
1704
1705         if (S_ISDIR(inode->i_mode)) {
1706                 int rc;
1707
1708                 rc = ll_update_lsm_md(inode, md);
1709                 if (rc)
1710                         return rc;
1711         }
1712
1713 #ifdef CONFIG_FS_POSIX_ACL
1714         if (body->mbo_valid & OBD_MD_FLACL) {
1715                 spin_lock(&lli->lli_lock);
1716                 if (lli->lli_posix_acl)
1717                         posix_acl_release(lli->lli_posix_acl);
1718                 lli->lli_posix_acl = md->posix_acl;
1719                 spin_unlock(&lli->lli_lock);
1720         }
1721 #endif
1722         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1723                                         sbi->ll_flags & LL_SBI_32BIT_API);
1724         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1725
1726         if (body->mbo_valid & OBD_MD_FLATIME) {
1727                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1728                         LTIME_S(inode->i_atime) = body->mbo_atime;
1729                 lli->lli_atime = body->mbo_atime;
1730         }
1731         if (body->mbo_valid & OBD_MD_FLMTIME) {
1732                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1733                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1734                                inode->i_ino, LTIME_S(inode->i_mtime),
1735                                body->mbo_mtime);
1736                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1737                 }
1738                 lli->lli_mtime = body->mbo_mtime;
1739         }
1740         if (body->mbo_valid & OBD_MD_FLCTIME) {
1741                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1742                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1743                 lli->lli_ctime = body->mbo_ctime;
1744         }
1745         if (body->mbo_valid & OBD_MD_FLMODE)
1746                 inode->i_mode = (inode->i_mode & S_IFMT) |
1747                                 (body->mbo_mode & ~S_IFMT);
1748         if (body->mbo_valid & OBD_MD_FLTYPE)
1749                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1750                                 (body->mbo_mode & S_IFMT);
1751         LASSERT(inode->i_mode != 0);
1752         if (S_ISREG(inode->i_mode))
1753                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1754                                        LL_MAX_BLKSIZE_BITS);
1755         else
1756                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1757         if (body->mbo_valid & OBD_MD_FLUID)
1758                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1759         if (body->mbo_valid & OBD_MD_FLGID)
1760                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1761         if (body->mbo_valid & OBD_MD_FLFLAGS)
1762                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1763         if (body->mbo_valid & OBD_MD_FLNLINK)
1764                 set_nlink(inode, body->mbo_nlink);
1765         if (body->mbo_valid & OBD_MD_FLRDEV)
1766                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1767
1768         if (body->mbo_valid & OBD_MD_FLID) {
1769                 /* FID shouldn't be changed! */
1770                 if (fid_is_sane(&lli->lli_fid)) {
1771                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1772                                  "Trying to change FID "DFID" to the "DFID", inode "DFID"(%p)\n",
1773                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1774                                  PFID(ll_inode2fid(inode)), inode);
1775                 } else {
1776                         lli->lli_fid = body->mbo_fid1;
1777                 }
1778         }
1779
1780         LASSERT(fid_seq(&lli->lli_fid) != 0);
1781
1782         if (body->mbo_valid & OBD_MD_FLSIZE) {
1783                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1784                     S_ISREG(inode->i_mode)) {
1785                         struct lustre_handle lockh;
1786                         enum ldlm_mode mode;
1787
1788                         /* As it is possible a blocking ast has been processed
1789                          * by this time, we need to check there is an UPDATE
1790                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1791                          * it.
1792                          */
1793                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1794                                                &lockh, LDLM_FL_CBPENDING,
1795                                                LCK_CR | LCK_CW |
1796                                                LCK_PR | LCK_PW);
1797                         if (mode) {
1798                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1799                                                       LLIF_EPOCH_PENDING |
1800                                                       LLIF_SOM_DIRTY)) {
1801                                         CERROR("%s: inode "DFID" flags %u still has size authority! do not trust the size got from MDS\n",
1802                                                sbi->ll_md_exp->exp_obd->obd_name,
1803                                                PFID(ll_inode2fid(inode)),
1804                                                lli->lli_flags);
1805                                 } else {
1806                                         /* Use old size assignment to avoid
1807                                          * deadlock bz14138 & bz14326
1808                                          */
1809                                         i_size_write(inode, body->mbo_size);
1810                                         spin_lock(&lli->lli_lock);
1811                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1812                                         spin_unlock(&lli->lli_lock);
1813                                 }
1814                                 ldlm_lock_decref(&lockh, mode);
1815                         }
1816                 } else {
1817                         /* Use old size assignment to avoid
1818                          * deadlock bz14138 & bz14326
1819                          */
1820                         i_size_write(inode, body->mbo_size);
1821
1822                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1823                                inode->i_ino, (unsigned long long)body->mbo_size);
1824                 }
1825
1826                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1827                         inode->i_blocks = body->mbo_blocks;
1828         }
1829
1830         if (body->mbo_valid & OBD_MD_TSTATE) {
1831                 if (body->mbo_t_state & MS_RESTORE)
1832                         lli->lli_flags |= LLIF_FILE_RESTORING;
1833         }
1834
1835         return 0;
1836 }
1837
1838 int ll_read_inode2(struct inode *inode, void *opaque)
1839 {
1840         struct lustre_md *md = opaque;
1841         struct ll_inode_info *lli = ll_i2info(inode);
1842         int rc;
1843
1844         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1845                PFID(&lli->lli_fid), inode);
1846
1847         LASSERT(!lli->lli_has_smd);
1848
1849         /* Core attributes from the MDS first.  This is a new inode, and
1850          * the VFS doesn't zero times in the core inode so we have to do
1851          * it ourselves.  They will be overwritten by either MDS or OST
1852          * attributes - we just need to make sure they aren't newer.
1853          */
1854         LTIME_S(inode->i_mtime) = 0;
1855         LTIME_S(inode->i_atime) = 0;
1856         LTIME_S(inode->i_ctime) = 0;
1857         inode->i_rdev = 0;
1858         rc = ll_update_inode(inode, md);
1859         if (rc)
1860                 return rc;
1861
1862         /* OIDEBUG(inode); */
1863
1864         if (S_ISREG(inode->i_mode)) {
1865                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1866
1867                 inode->i_op = &ll_file_inode_operations;
1868                 inode->i_fop = sbi->ll_fop;
1869                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1870         } else if (S_ISDIR(inode->i_mode)) {
1871                 inode->i_op = &ll_dir_inode_operations;
1872                 inode->i_fop = &ll_dir_operations;
1873         } else if (S_ISLNK(inode->i_mode)) {
1874                 inode->i_op = &ll_fast_symlink_inode_operations;
1875         } else {
1876                 inode->i_op = &ll_special_inode_operations;
1877
1878                 init_special_inode(inode, inode->i_mode,
1879                                    inode->i_rdev);
1880         }
1881
1882         return 0;
1883 }
1884
1885 void ll_delete_inode(struct inode *inode)
1886 {
1887         struct ll_inode_info *lli = ll_i2info(inode);
1888
1889         if (S_ISREG(inode->i_mode) && lli->lli_clob)
1890                 /* discard all dirty pages before truncating them, required by
1891                  * osc_extent implementation at LU-1030.
1892                  */
1893                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1894                                    CL_FSYNC_DISCARD, 1);
1895
1896         truncate_inode_pages_final(&inode->i_data);
1897
1898         /* Workaround for LU-118 */
1899         if (inode->i_data.nrpages) {
1900                 spin_lock_irq(&inode->i_data.tree_lock);
1901                 spin_unlock_irq(&inode->i_data.tree_lock);
1902                 LASSERTF(inode->i_data.nrpages == 0,
1903                          "inode="DFID"(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1904                          PFID(ll_inode2fid(inode)), inode,
1905                          inode->i_data.nrpages);
1906         }
1907         /* Workaround end */
1908
1909         ll_clear_inode(inode);
1910         clear_inode(inode);
1911 }
1912
1913 int ll_iocontrol(struct inode *inode, struct file *file,
1914                  unsigned int cmd, unsigned long arg)
1915 {
1916         struct ll_sb_info *sbi = ll_i2sbi(inode);
1917         struct ptlrpc_request *req = NULL;
1918         int rc, flags = 0;
1919
1920         switch (cmd) {
1921         case FSFILT_IOC_GETFLAGS: {
1922                 struct mdt_body *body;
1923                 struct md_op_data *op_data;
1924
1925                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1926                                              0, 0, LUSTRE_OPC_ANY,
1927                                              NULL);
1928                 if (IS_ERR(op_data))
1929                         return PTR_ERR(op_data);
1930
1931                 op_data->op_valid = OBD_MD_FLFLAGS;
1932                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1933                 ll_finish_md_op_data(op_data);
1934                 if (rc) {
1935                         CERROR("%s: failure inode "DFID": rc = %d\n",
1936                                sbi->ll_md_exp->exp_obd->obd_name,
1937                                PFID(ll_inode2fid(inode)), rc);
1938                         return -abs(rc);
1939                 }
1940
1941                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1942
1943                 flags = body->mbo_flags;
1944
1945                 ptlrpc_req_finished(req);
1946
1947                 return put_user(flags, (int __user *)arg);
1948         }
1949         case FSFILT_IOC_SETFLAGS: {
1950                 struct lov_stripe_md *lsm;
1951                 struct obd_info oinfo = { };
1952                 struct md_op_data *op_data;
1953
1954                 if (get_user(flags, (int __user *)arg))
1955                         return -EFAULT;
1956
1957                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1958                                              LUSTRE_OPC_ANY, NULL);
1959                 if (IS_ERR(op_data))
1960                         return PTR_ERR(op_data);
1961
1962                 op_data->op_attr_flags = flags;
1963                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1964                 rc = md_setattr(sbi->ll_md_exp, op_data,
1965                                 NULL, 0, NULL, 0, &req, NULL);
1966                 ll_finish_md_op_data(op_data);
1967                 ptlrpc_req_finished(req);
1968                 if (rc)
1969                         return rc;
1970
1971                 inode->i_flags = ll_ext_to_inode_flags(flags);
1972
1973                 lsm = ccc_inode_lsm_get(inode);
1974                 if (!lsm_has_objects(lsm)) {
1975                         ccc_inode_lsm_put(inode, lsm);
1976                         return 0;
1977                 }
1978
1979                 oinfo.oi_oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
1980                 if (!oinfo.oi_oa) {
1981                         ccc_inode_lsm_put(inode, lsm);
1982                         return -ENOMEM;
1983                 }
1984                 oinfo.oi_md = lsm;
1985                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1986                 oinfo.oi_oa->o_flags = flags;
1987                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1988                                        OBD_MD_FLGROUP;
1989                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1990                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1991                 kmem_cache_free(obdo_cachep, oinfo.oi_oa);
1992                 ccc_inode_lsm_put(inode, lsm);
1993
1994                 if (rc && rc != -EPERM && rc != -EACCES)
1995                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1996
1997                 return rc;
1998         }
1999         default:
2000                 return -ENOSYS;
2001         }
2002
2003         return 0;
2004 }
2005
2006 int ll_flush_ctx(struct inode *inode)
2007 {
2008         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2009
2010         CDEBUG(D_SEC, "flush context for user %d\n",
2011                from_kuid(&init_user_ns, current_uid()));
2012
2013         obd_set_info_async(NULL, sbi->ll_md_exp,
2014                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2015                            0, NULL, NULL);
2016         obd_set_info_async(NULL, sbi->ll_dt_exp,
2017                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2018                            0, NULL, NULL);
2019         return 0;
2020 }
2021
2022 /* umount -f client means force down, don't save state */
2023 void ll_umount_begin(struct super_block *sb)
2024 {
2025         struct ll_sb_info *sbi = ll_s2sbi(sb);
2026         struct obd_device *obd;
2027         struct obd_ioctl_data *ioc_data;
2028
2029         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2030                sb->s_count, atomic_read(&sb->s_active));
2031
2032         obd = class_exp2obd(sbi->ll_md_exp);
2033         if (!obd) {
2034                 CERROR("Invalid MDC connection handle %#llx\n",
2035                        sbi->ll_md_exp->exp_handle.h_cookie);
2036                 return;
2037         }
2038         obd->obd_force = 1;
2039
2040         obd = class_exp2obd(sbi->ll_dt_exp);
2041         if (!obd) {
2042                 CERROR("Invalid LOV connection handle %#llx\n",
2043                        sbi->ll_dt_exp->exp_handle.h_cookie);
2044                 return;
2045         }
2046         obd->obd_force = 1;
2047
2048         ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
2049         if (ioc_data) {
2050                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2051                               sizeof(*ioc_data), ioc_data, NULL);
2052
2053                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2054                               sizeof(*ioc_data), ioc_data, NULL);
2055
2056                 kfree(ioc_data);
2057         }
2058
2059         /* Really, we'd like to wait until there are no requests outstanding,
2060          * and then continue.  For now, we just invalidate the requests,
2061          * schedule() and sleep one second if needed, and hope.
2062          */
2063         schedule();
2064 }
2065
2066 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2067 {
2068         struct ll_sb_info *sbi = ll_s2sbi(sb);
2069         char *profilenm = get_profile_name(sb);
2070         int err;
2071         __u32 read_only;
2072
2073         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2074                 read_only = *flags & MS_RDONLY;
2075                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2076                                          sizeof(KEY_READ_ONLY),
2077                                          KEY_READ_ONLY, sizeof(read_only),
2078                                          &read_only, NULL);
2079                 if (err) {
2080                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2081                                       profilenm, read_only ?
2082                                       "read-only" : "read-write", err);
2083                         return err;
2084                 }
2085
2086                 if (read_only)
2087                         sb->s_flags |= MS_RDONLY;
2088                 else
2089                         sb->s_flags &= ~MS_RDONLY;
2090
2091                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2092                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2093                                       read_only ?  "read-only" : "read-write");
2094         }
2095         return 0;
2096 }
2097
2098 /**
2099  * Cleanup the open handle that is cached on MDT-side.
2100  *
2101  * For open case, the client side open handling thread may hit error
2102  * after the MDT grant the open. Under such case, the client should
2103  * send close RPC to the MDT as cleanup; otherwise, the open handle
2104  * on the MDT will be leaked there until the client umount or evicted.
2105  *
2106  * In further, if someone unlinked the file, because the open handle
2107  * holds the reference on such file/object, then it will block the
2108  * subsequent threads that want to locate such object via FID.
2109  *
2110  * \param[in] sb        super block for this file-system
2111  * \param[in] open_req  pointer to the original open request
2112  */
2113 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2114 {
2115         struct mdt_body                 *body;
2116         struct md_op_data               *op_data;
2117         struct ptlrpc_request           *close_req = NULL;
2118         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2119
2120         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2121         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2122         if (!op_data)
2123                 return;
2124
2125         op_data->op_fid1 = body->mbo_fid1;
2126         op_data->op_ioepoch = body->mbo_ioepoch;
2127         op_data->op_handle = body->mbo_handle;
2128         op_data->op_mod_time = get_seconds();
2129         md_close(exp, op_data, NULL, &close_req);
2130         ptlrpc_req_finished(close_req);
2131         ll_finish_md_op_data(op_data);
2132 }
2133
2134 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2135                   struct super_block *sb, struct lookup_intent *it)
2136 {
2137         struct ll_sb_info *sbi = NULL;
2138         struct lustre_md md = { NULL };
2139         int rc;
2140
2141         LASSERT(*inode || sb);
2142         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2143         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2144                               sbi->ll_md_exp, &md);
2145         if (rc)
2146                 goto cleanup;
2147
2148         if (*inode) {
2149                 rc = ll_update_inode(*inode, &md);
2150                 if (rc)
2151                         goto out;
2152         } else {
2153                 LASSERT(sb);
2154
2155                 /*
2156                  * At this point server returns to client's same fid as client
2157                  * generated for creating. So using ->fid1 is okay here.
2158                  */
2159                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2160                         CERROR("%s: Fid is insane " DFID "\n",
2161                                ll_get_fsname(sb, NULL, 0),
2162                                PFID(&md.body->mbo_fid1));
2163                         rc = -EINVAL;
2164                         goto out;
2165                 }
2166
2167                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2168                                              sbi->ll_flags & LL_SBI_32BIT_API),
2169                                  &md);
2170                 if (IS_ERR(*inode)) {
2171 #ifdef CONFIG_FS_POSIX_ACL
2172                         if (md.posix_acl) {
2173                                 posix_acl_release(md.posix_acl);
2174                                 md.posix_acl = NULL;
2175                         }
2176 #endif
2177                         rc = -ENOMEM;
2178                         CERROR("new_inode -fatal: rc %d\n", rc);
2179                         goto out;
2180                 }
2181         }
2182
2183         /* Handling piggyback layout lock.
2184          * Layout lock can be piggybacked by getattr and open request.
2185          * The lsm can be applied to inode only if it comes with a layout lock
2186          * otherwise correct layout may be overwritten, for example:
2187          * 1. proc1: mdt returns a lsm but not granting layout
2188          * 2. layout was changed by another client
2189          * 3. proc2: refresh layout and layout lock granted
2190          * 4. proc1: to apply a stale layout
2191          */
2192         if (it && it->it_lock_mode != 0) {
2193                 struct lustre_handle lockh;
2194                 struct ldlm_lock *lock;
2195
2196                 lockh.cookie = it->it_lock_handle;
2197                 lock = ldlm_handle2lock(&lockh);
2198                 LASSERT(lock);
2199                 if (ldlm_has_layout(lock)) {
2200                         struct cl_object_conf conf;
2201
2202                         memset(&conf, 0, sizeof(conf));
2203                         conf.coc_opc = OBJECT_CONF_SET;
2204                         conf.coc_inode = *inode;
2205                         conf.coc_lock = lock;
2206                         conf.u.coc_md = &md;
2207                         (void)ll_layout_conf(*inode, &conf);
2208                 }
2209                 LDLM_LOCK_PUT(lock);
2210         }
2211
2212 out:
2213         if (md.lsm)
2214                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2215         md_free_lustre_md(sbi->ll_md_exp, &md);
2216
2217 cleanup:
2218         if (rc != 0 && it && it->it_op & IT_OPEN)
2219                 ll_open_cleanup(sb ? sb : (*inode)->i_sb, req);
2220
2221         return rc;
2222 }
2223
2224 int ll_obd_statfs(struct inode *inode, void __user *arg)
2225 {
2226         struct ll_sb_info *sbi = NULL;
2227         struct obd_export *exp;
2228         char *buf = NULL;
2229         struct obd_ioctl_data *data = NULL;
2230         __u32 type;
2231         int len = 0, rc;
2232
2233         if (!inode) {
2234                 rc = -EINVAL;
2235                 goto out_statfs;
2236         }
2237
2238         sbi = ll_i2sbi(inode);
2239         if (!sbi) {
2240                 rc = -EINVAL;
2241                 goto out_statfs;
2242         }
2243
2244         rc = obd_ioctl_getdata(&buf, &len, arg);
2245         if (rc)
2246                 goto out_statfs;
2247
2248         data = (void *)buf;
2249         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2250             !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2251                 rc = -EINVAL;
2252                 goto out_statfs;
2253         }
2254
2255         if (data->ioc_inllen1 != sizeof(__u32) ||
2256             data->ioc_inllen2 != sizeof(__u32) ||
2257             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2258             data->ioc_plen2 != sizeof(struct obd_uuid)) {
2259                 rc = -EINVAL;
2260                 goto out_statfs;
2261         }
2262
2263         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2264         if (type & LL_STATFS_LMV) {
2265                 exp = sbi->ll_md_exp;
2266         } else if (type & LL_STATFS_LOV) {
2267                 exp = sbi->ll_dt_exp;
2268         } else {
2269                 rc = -ENODEV;
2270                 goto out_statfs;
2271         }
2272
2273         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2274         if (rc)
2275                 goto out_statfs;
2276 out_statfs:
2277         if (buf)
2278                 obd_ioctl_freedata(buf, len);
2279         return rc;
2280 }
2281
2282 int ll_process_config(struct lustre_cfg *lcfg)
2283 {
2284         char *ptr;
2285         void *sb;
2286         struct lprocfs_static_vars lvars;
2287         unsigned long x;
2288         int rc = 0;
2289
2290         lprocfs_llite_init_vars(&lvars);
2291
2292         /* The instance name contains the sb: lustre-client-aacfe000 */
2293         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2294         if (!ptr || !*(++ptr))
2295                 return -EINVAL;
2296         rc = kstrtoul(ptr, 16, &x);
2297         if (rc != 0)
2298                 return -EINVAL;
2299         sb = (void *)x;
2300         /* This better be a real Lustre superblock! */
2301         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2302
2303         /* Note we have not called client_common_fill_super yet, so
2304          * proc fns must be able to handle that!
2305          */
2306         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2307                                       lcfg, sb);
2308         if (rc > 0)
2309                 rc = 0;
2310         return rc;
2311 }
2312
2313 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2314 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2315                                       struct inode *i1, struct inode *i2,
2316                                       const char *name, int namelen,
2317                                       int mode, __u32 opc, void *data)
2318 {
2319         if (!name) {
2320                 /* Do not reuse namelen for something else. */
2321                 if (namelen)
2322                         return ERR_PTR(-EINVAL);
2323         } else {
2324                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2325                         return ERR_PTR(-ENAMETOOLONG);
2326
2327                 if (!lu_name_is_valid_2(name, namelen))
2328                         return ERR_PTR(-EINVAL);
2329         }
2330
2331         if (!op_data)
2332                 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2333
2334         if (!op_data)
2335                 return ERR_PTR(-ENOMEM);
2336
2337         ll_i2gids(op_data->op_suppgids, i1, i2);
2338         op_data->op_fid1 = *ll_inode2fid(i1);
2339         if (S_ISDIR(i1->i_mode))
2340                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2341
2342         if (i2) {
2343                 op_data->op_fid2 = *ll_inode2fid(i2);
2344                 if (S_ISDIR(i2->i_mode))
2345                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2346         } else {
2347                 fid_zero(&op_data->op_fid2);
2348         }
2349
2350         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2351                 op_data->op_cli_flags |= CLI_HASH64;
2352
2353         if (ll_need_32bit_api(ll_i2sbi(i1)))
2354                 op_data->op_cli_flags |= CLI_API32;
2355
2356         op_data->op_name = name;
2357         op_data->op_namelen = namelen;
2358         op_data->op_mode = mode;
2359         op_data->op_mod_time = ktime_get_real_seconds();
2360         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2361         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2362         op_data->op_cap = cfs_curproc_cap_pack();
2363         op_data->op_bias = 0;
2364         op_data->op_cli_flags = 0;
2365         if ((opc == LUSTRE_OPC_CREATE) && name &&
2366             filename_is_volatile(name, namelen, NULL))
2367                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2368         op_data->op_mds = 0;
2369         op_data->op_data = data;
2370
2371         /* When called by ll_setattr_raw, file is i1. */
2372         if (ll_i2info(i1)->lli_flags & LLIF_DATA_MODIFIED)
2373                 op_data->op_bias |= MDS_DATA_MODIFIED;
2374
2375         return op_data;
2376 }
2377
2378 void ll_finish_md_op_data(struct md_op_data *op_data)
2379 {
2380         kfree(op_data);
2381 }
2382
2383 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2384 {
2385         struct ll_sb_info *sbi;
2386
2387         LASSERT(seq && dentry);
2388         sbi = ll_s2sbi(dentry->d_sb);
2389
2390         if (sbi->ll_flags & LL_SBI_NOLCK)
2391                 seq_puts(seq, ",nolock");
2392
2393         if (sbi->ll_flags & LL_SBI_FLOCK)
2394                 seq_puts(seq, ",flock");
2395
2396         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2397                 seq_puts(seq, ",localflock");
2398
2399         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2400                 seq_puts(seq, ",user_xattr");
2401
2402         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2403                 seq_puts(seq, ",lazystatfs");
2404
2405         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2406                 seq_puts(seq, ",user_fid2path");
2407
2408         return 0;
2409 }
2410
2411 /**
2412  * Get obd name by cmd, and copy out to user space
2413  */
2414 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2415 {
2416         struct ll_sb_info *sbi = ll_i2sbi(inode);
2417         struct obd_device *obd;
2418
2419         if (cmd == OBD_IOC_GETDTNAME)
2420                 obd = class_exp2obd(sbi->ll_dt_exp);
2421         else if (cmd == OBD_IOC_GETMDNAME)
2422                 obd = class_exp2obd(sbi->ll_md_exp);
2423         else
2424                 return -EINVAL;
2425
2426         if (!obd)
2427                 return -ENOENT;
2428
2429         if (copy_to_user((void __user *)arg, obd->obd_name,
2430                          strlen(obd->obd_name) + 1))
2431                 return -EFAULT;
2432
2433         return 0;
2434 }
2435
2436 /**
2437  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2438  * fsname will be returned in this buffer; otherwise, a static buffer will be
2439  * used to store the fsname and returned to caller.
2440  */
2441 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2442 {
2443         static char fsname_static[MTI_NAME_MAXLEN];
2444         struct lustre_sb_info *lsi = s2lsi(sb);
2445         char *ptr;
2446         int len;
2447
2448         if (!buf) {
2449                 /* this means the caller wants to use static buffer
2450                  * and it doesn't care about race. Usually this is
2451                  * in error reporting path
2452                  */
2453                 buf = fsname_static;
2454                 buflen = sizeof(fsname_static);
2455         }
2456
2457         len = strlen(lsi->lsi_lmd->lmd_profile);
2458         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2459         if (ptr && (strcmp(ptr, "-client") == 0))
2460                 len -= 7;
2461
2462         if (unlikely(len >= buflen))
2463                 len = buflen - 1;
2464         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2465         buf[len] = '\0';
2466
2467         return buf;
2468 }
2469
2470 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2471 {
2472         char *buf, *path = NULL;
2473         struct dentry *dentry = NULL;
2474         struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2475
2476         /* this can be called inside spin lock so use GFP_ATOMIC. */
2477         buf = (char *)__get_free_page(GFP_ATOMIC);
2478         if (buf) {
2479                 dentry = d_find_alias(page->mapping->host);
2480                 if (dentry)
2481                         path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2482         }
2483
2484         CDEBUG(D_WARNING,
2485                "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2486                ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2487                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2488                PFID(&obj->vob_header.coh_lu.loh_fid),
2489                (path && !IS_ERR(path)) ? path : "", ioret);
2490
2491         if (dentry)
2492                 dput(dentry);
2493
2494         if (buf)
2495                 free_page((unsigned long)buf);
2496 }
2497
2498 /*
2499  * Compute llite root squash state after a change of root squash
2500  * configuration setting or add/remove of a lnet nid
2501  */
2502 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2503 {
2504         struct root_squash_info *squash = &sbi->ll_squash;
2505         lnet_process_id_t id;
2506         bool matched;
2507         int i;
2508
2509         /* Update norootsquash flag */
2510         down_write(&squash->rsi_sem);
2511         if (list_empty(&squash->rsi_nosquash_nids)) {
2512                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2513         } else {
2514                 /*
2515                  * Do not apply root squash as soon as one of our NIDs is
2516                  * in the nosquash_nids list
2517                  */
2518                 matched = false;
2519                 i = 0;
2520
2521                 while (LNetGetId(i++, &id) != -ENOENT) {
2522                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2523                                 continue;
2524                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2525                                 matched = true;
2526                                 break;
2527                         }
2528                 }
2529                 if (matched)
2530                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2531                 else
2532                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2533         }
2534         up_write(&squash->rsi_sem);
2535 }